CN111046783B - Slope geological disaster boundary extraction method for improving watershed algorithm - Google Patents
Slope geological disaster boundary extraction method for improving watershed algorithm Download PDFInfo
- Publication number
- CN111046783B CN111046783B CN201911250891.5A CN201911250891A CN111046783B CN 111046783 B CN111046783 B CN 111046783B CN 201911250891 A CN201911250891 A CN 201911250891A CN 111046783 B CN111046783 B CN 111046783B
- Authority
- CN
- China
- Prior art keywords
- region
- value
- luv
- values
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Abstract
The invention belongs to the field of geological disaster information extraction, and discloses a slope geological disaster boundary extraction method for improving a watershed algorithm, which comprises the following steps of: s1, dividing an image to be divided by a simulated immersion watershed algorithm; converting the RGB values of the pixels in each region obtained after the segmentation into Luv values, and calculating the Luv average value of all the pixels in each region to be used as the Luv value of the region; s2, establishing four neighborhood arrays of each region with the Luv mean value as the region color, and simultaneously determining a minimum region judgment threshold value; s3, finding all the minimum areas; s4, merging the adjacent areas with the minimum areas by a color difference judging method until all the minimum areas are merged; s5, converting the Luv values of all the combined areas into RGB values. The invention has obvious advantages in information extraction time, avoids complex process of post-treatment and greatly improves time efficiency.
Description
Technical Field
The invention belongs to the field of geological disaster information extraction, and particularly relates to a slope geological disaster boundary extraction method for improving a watershed algorithm.
Background
Slope geological disasters are one of common geological disasters and greatly influence the surrounding environment of a disaster body and the life and property safety of people. The development of remote sensing technology provides a faster means for extracting slope disaster information, and at present, although a plurality of slope disaster extraction researches based on image classification and image segmentation methods exist, the slope disaster information extracted through remote sensing images still stays in manual visual interpretation based on GIS software in production. The method not only requires technicians to have rich ground learning knowledge and interpretation experience, but also requires a great deal of manpower and time investment, has low production efficiency, and the extracted slope disaster information has larger subjectivity and uncertainty, so that the application requirements of post-disaster emergency investigation, disaster assessment and the like are difficult to meet. Under the support of rapid development of the high-resolution remote sensing image segmentation technology, the development of automatic slope disaster boundary extraction of the high-resolution remote sensing image is increasingly feasible.
Among the image segmentation technologies, the watershed image segmentation method is one of the common image segmentation methods, the segmentation result is a single-pixel closed and communicated region, and meanwhile, the contour line and the segmented object have good matching degree, so that the watershed image segmentation method can be used as an image segmentation method for slope geological disaster boundary extraction. The watershed segmentation algorithm is based on the color difference of image pixels, namely the more obvious the contrast is, the better the segmentation effect is. The slope area of the high-resolution remote sensing image is an area with higher texture and spectrum consistency, and the tone is usually different from the tone of surrounding land, so that the precondition of automatic extraction and application of the slope disaster boundary is provided for the watershed image segmentation technology.
At present, the watershed segmentation algorithm is continuously and deeply researched, from the proposal of an overflow method to the application, segmented images are developed from gray images to color image segmentation, the watershed segmentation is performed with higher algorithm speed, the watershed segmentation is performed based on texture and morphological gradient fusion, and until 2003, the watershed segmentation algorithm is systematically summarized by Soille, so that the watershed algorithm has numerous applications in the field of remote sensing image information extraction. Aiming at the problems of image over-segmentation, obvious algorithm noise and the like of the watershed algorithm, a plurality of improved watershed segmentation algorithms are provided. The watershed segmentation algorithm based on region merging can realize segmented region merging based on region texture features and the like after segmentation is completed, and the research of image brightness balance is developed based on Lab color space color difference distance measurement and gradually developed into watershed image segmented region merging research, so that good segmentation results are obtained. Meanwhile, an RGB color model is generally used in the conventional image color expression, but in this color model, R, G, B three components have a high correlation, that is, changing the brightness of an image, any one of the three components may be changed, and this color definition property is inappropriate for image segmentation.
Thus, there is a need for improvements to existing watershed algorithms to adapt them for use in slope geologic hazard boundary extraction.
Disclosure of Invention
The invention overcomes the defects existing in the prior art, and solves the technical problems that: the slope geological disaster boundary extraction method for improving the watershed algorithm is provided to realize automatic and accurate extraction of the slope disaster boundary.
In order to solve the technical problems, the invention adopts the following technical scheme: a slope geological disaster boundary extraction method for improving a watershed algorithm comprises the following steps:
s1, dividing an image to be divided by a simulated immersion watershed algorithm; converting the RGB values of the pixels in each region obtained after the segmentation into Luv values, and calculating the Luv average value of all the pixels in each region to be used as the Luv value of the region;
s2, establishing four neighborhood arrays of each region with the Luv mean value as the region color, and simultaneously determining a minimum region judgment threshold value;
s3, scanning all the areas in sequence, judging whether the total number of pixels of the areas is smaller than a minimum area judging threshold value, and if so, classifying the pixels into minimum areas until all the minimum areas are found;
s4, traversing all adjacent areas of each determined minimum area, and calculating color difference values d of the minimum area and all the adjacent areas according to the Luv mean value of the adjacent areas i Satisfying the color difference value
Combining the region of (2) with the very small region; a new region is formed after combination, information of all adjacent regions of the new region after combination is refreshed, the mean value of the Luv values of the two regions before combination is used as the Luv value of the new region, and D is a color difference value threshold;
s5, judging the pixel value of the new combined region, judging whether the new combined region is still a minimum region, and if so, returning to the step S4 to be combined again; if not, judging whether other minimum areas exist, if so, returning to the step S4 to continue merging until all the minimum areas are merged;
s6, converting the Luv values of all the combined areas into RGB values.
In the step S2, the minimum area determination threshold a min The values of (2) are:
A min ＝(M×N)/C；
wherein M is the row value of the image to be segmented, N is the column value of the image to be segmented, and C is a constant.
In the slope geological disaster boundary extraction method of the improved watershed algorithm, the value of the constant C is 500, and the value D of the color difference value threshold is 400.
In the step S4, the color difference value d i The calculation formula of (2) is as follows:
wherein R is i |、|R j I respectively denote adjacent regions R i And a minimum region R j The number of pixels included in F c (R i )、F c (R j ) Respectively represent adjacent regions R i And a minimum region R j And n is the number of adjacent areas.
In the step S1, the image to be segmented is segmented by a watershed algorithm, which specifically includes the following steps:
(1) Converting pixel values of the image to be segmented to convert RGB values into gray values;
(2) Calculating to obtain gradients of each pixel point in the horizontal and vertical directions, and counting the frequency and the accumulation probability of each gradient;
(3) Sequencing according to the gradient values, determining the positions of the gradient values in a sequencing array, wherein the same gradient is in the same gradient level;
(4) Processing all pixel points of the first gradient level, checking whether the neighborhood of the point belongs to a certain region or watershed, if yes, adding the point into a first-in first-out queue;
(5) Starting to expand the existing basin according to the first-in first-out queue, scanning the pixel neighborhood in the first-in first-out queue, if the gradients are equal, namely the same gradient level is not a watershed, refreshing the identification of the pixel by using the identification of the neighborhood pixel, and circularly completing the expansion of all pixel points in the queue;
(6) Judging whether the pixel points are not identified, if so, continuing to execute the step (5) on the pixel points until all the pixel points in the queue are expanded;
(7) After the first gradient level is processed, returning to the step (4) to continuously process the next gradient level, and circulating until all gradient levels are processed, so as to obtain a water distribution line of the gradient image, namely a boundary line of image segmentation;
(8) The image is divided into a large number of areas by the dividing boundary line, and the areas are the image dividing results and represent different types of information.
In the step S1, the process of converting the RGB values of the pixels in each region into the Luv values is implemented based on color space conversion, and the conversion relationship is as follows:
u＝13L(u'-u n ')；
v＝13L(v'-v n ')；
wherein:
u'＝4X/(X+15Y+3Z)；
v'＝9Y/(X+15Y+3Z)；
u n '＝4X n /(X n +15Y n +3Z n )；
v n '＝9Y n /(X n +15Y n +3Z n )；
wherein the L value represents the brightness of the pixel, the value range is 0-100, the u and v values represent the chromaticity coordinates, the value range is-100, and X n ，Y n ，Z n Representing the coordinates of the CIE standard illuminant, X, Y, Z representing the values of the CIE XYZ color space.
Compared with the prior art, the invention has the following beneficial effects:
(1) The invention provides a slope geological disaster boundary extraction method for improving a watershed algorithm, which has obvious advantages in information extraction time through experimental verification, avoids the complex process of post-treatment and greatly improves the time efficiency. Meanwhile, the method automatically combines the initial segmentation results, avoids subjectivity of manually combining and processing broken spots after classification, and has good comprehensive extraction efficiency and objectivity of the extraction results.
(2) The invention does not need to establish a segmentation process rule, has simple merging process, easy algorithm understanding, strong objectivity of segmentation result and high reliability. The multi-scale segmentation test result of the unstable slope boundary of the test image shows that the slope geological disaster boundary extraction method provided by the invention has good extraction reliability, matching degree of the boundary of the target body and extraction details. Meanwhile, the calculation result of the segmentation accuracy evaluation criterion shows that the extraction result of the invention has higher accuracy, and the evaluation result is consistent with the visual evaluation result, which indicates that the segmentation accuracy evaluation criterion factor result of the watershed algorithm image used by the invention is reliable.
Drawings
FIG. 1 is a diagram of geographic locations of an experimental area employed in an embodiment of the present invention;
FIG. 2 is a schematic diagram of a GF-2 remote sensing image of an experimental region according to an embodiment of the present invention, wherein A represents an original GF-2 image; b represents the GF-2 image after contrast enhancement;
FIG. 3 is a panoramic view of an unstable ramp body of an experimental zone in an embodiment of the present invention;
FIG. 4 is a schematic diagram of an experimental technique according to an embodiment of the present invention;
FIG. 5 is a graph showing the result of a slope disaster boundary segmentation test using the RGB-RMWS method according to an embodiment of the invention, wherein A represents an original image; b represents the image after contrast enhancement;
FIG. 6 is a graph showing the result of a Luv-RMWS method slope disaster boundary segmentation test, wherein A represents an original image; b represents the image after contrast enhancement;
FIG. 7 is a comparison of the results of a multi-scale Luv-RMWS ramp disaster boundary segmentation experiment in an embodiment of the invention;
FIG. 8 is a graph showing the comparison of the unstable ramp boundary extraction result with the reference data, wherein A is the result of RGB-RMWS method; b is the result of the Luv-RMWS method.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments; all other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The embodiment of the invention provides a slope geological disaster boundary extraction method for improving a watershed algorithm, which comprises the following steps of:
s1, dividing an image to be divided by a simulated immersion watershed algorithm; the RGB values of the pixels in each region obtained by the division are converted into Luv values, and the Luv average value of all the pixels in each region is obtained as the Luv value of the region.
In the step S1, the image to be segmented is segmented by a watershed algorithm, which specifically includes the following steps:
(1) Converting pixel values of the image to be segmented to convert RGB values into gray values;
(2) Calculating to obtain gradients of each pixel point in the horizontal and vertical directions, and counting the frequency and the accumulation probability of each gradient;
the gradient of the edge pixel is 0, and the value range of the gradient is as follows: 0-255 (255 being replaced with 255 being greater than 255 and 0 being replaced with 0 being less than 0). The gradient function is:
wherein f (x, y) is the original image, and G () is the gradient operation.
(3) Sequencing according to the gradient values, determining the positions of the gradient values in a sequencing array, wherein the same gradient is in the same gradient level;
(4) Processing all pixel points of the first gradient level, checking whether the neighborhood of the point belongs to a certain region or watershed, if yes, adding the point into a first-in first-out queue;
(5) Starting to expand the existing basin according to the first-in first-out queue, scanning the pixel neighborhood in the first-in first-out queue, if the gradients are equal, namely the same gradient level is not a watershed, refreshing the identification of the pixel by using the identification of the neighborhood pixel, and circularly completing the expansion of all pixel points in the queue;
(6) Judging whether the pixel points are not identified, if so, continuing to execute the step (5) on the pixel points until all the pixel points in the queue are expanded;
(7) After the first gradient level is processed, returning to the step (4) to continuously process the next gradient level, and circulating until all gradient levels are processed, so as to obtain a water distribution line of the gradient image, namely a boundary line of image segmentation;
(8) The image is divided into a large number of areas by the dividing boundary line, and the areas are the image dividing results and represent different types of information.
In addition, in step S1, the process of converting the RGB values of the pixels of each region into Luv values is implemented based on color space conversion, and the conversion relationship thereof is as follows:
u＝13L(u'-u n ')； (4)
v＝13L(v'-v n ')； (5)
wherein, the liquid crystal display device comprises a liquid crystal display device,
u'＝4X/(X+15Y+3Z)； (6)
v'＝9Y/(X+15Y+3Z)； (7)
u n '＝4X n /(X n +15Y n +3Z n )； (8)
v n '＝9Y n /(X n +15Y n +3Z n )； (9)
wherein the L value represents the brightness of the pixel, the value range is 0-100, the u and v values represent the chromaticity coordinates, the value range is-100, and the u n ' and v n ' represents the coordinates of the CIE standard illuminant, and is a tristimulus value. X is X n ，Y n ，Z n Also expressed as coordinates of CIE standard light source, is tristimulus value, and is generally taken as:0.9505, 1.0000, 1.0888.X, Y, Z is the value of the CIE XYZ color space for converting the RGB color space to the Luv color space.
In the case of a 2 ° observer and a C light source, u n '＝0.2009，v n '＝0.4610。
In the Luv color space, the difference between any two colors is called chromatic aberration. The color difference is the distance between the color positions, denoted by Δe, i.e. the color difference between two colors is calculated as follows:
ΔE＝(ΔL 2 +Δu 2 +Δv 2 ) 1/2 ； (10)
where Δl represents the luminance difference, and Δa and Δb represent the difference between the two colors in the u and v directions.
S2, establishing four neighborhood arrays of each region with the Luv mean value as the region color, and simultaneously determining a minimum region judgment threshold value;
since the watershed algorithm divides the scale parameter into substantially the determination threshold value of the minimum region, in this embodiment, the minimum region determination threshold value a min The values of (2) are:
A min ＝(M×N)/C； (11)
wherein M is the row value of the image to be segmented, N is the column value of the image to be segmented, and C is a constant.
Clearly minimum area decision threshold A min For images with different sizes and different values, instead of a fixed amount, the optimal segmentation scale parameter A can be determined by repeating experiments through trial and error min But it is essential to determine the constant value C.
S3, scanning all the areas in sequence, judging whether the total number of pixels of the areas is smaller than a minimum area judging threshold value, and if so, classifying the pixels into the minimum areas until all the minimum areas are found.
S4, traversing all adjacent areas of each determined minimum area, and calculating color difference values d of the minimum area and all the adjacent areas according to the Luv mean value of the adjacent areas i Satisfying the color difference value
Is combined with the region of the smallest area, +.>
The regions of (2) are not merged; and after merging, forming a new region, refreshing information of all adjacent regions of the new region after merging, taking the mean value of the Luv values of the two regions before merging as the Luv value of the new region, wherein D is a color difference value threshold.
In the step S4, the color difference value d i The calculation formula of (2) is as follows:
wherein R is i |、|R j I respectively denote adjacent regions R i And a minimum region R j The number of pixels included in F c (R i )、F c (R j ) Respectively represent adjacent regions R i And a minimum region R j And n is the number of adjacent areas.
And adopting chromatic aberration to judge similarity measures of the current minimum area and all adjacent areas. When d i At 1 or less, the two regions are not distinguishable in color, i.e. d i The smaller the two regions are, the more similar the color. In the region merging, whether the colors of adjacent regions are similar needs to be judged, so d needs to be determined through theoretical analysis or empirical verification i If the color difference threshold is D, the region merging of the segmentation result is completed by the constraint of D until no similar region merging exists. The color difference threshold D may also be determined by trial and error.
S5, judging the pixel value of the new combined region, judging whether the new combined region is still a minimum region, and if so, returning to the step S4 to be combined again; if not, judging whether other minimum areas exist, if so, returning to the step S4 to continue merging until all the minimum areas are merged.
S6, converting the Luv values of all the combined areas into RGB values.
In order to achieve a better display effect, the Luv values of all the combined areas are converted into RGB values, and the final segmentation result area of the image is displayed in the RGB values.
The quality of the image segmentation effect directly influences the result and the precision of the subsequent information analysis processing, so that the comprehensive and objective evaluation of the remote sensing image segmentation method is necessary, the image segmentation precision evaluation is as important as the image segmentation technology, and the evaluation is often carried out in a qualitative and quantitative mode. However, there are various uncertain factors in the remote sensing image segmentation, and quantitative evaluation of the advantages and disadvantages of different image segmentation algorithms is one of the recognized difficulties in the image segmentation research field. The most commonly used segmentation precision evaluation method is still a subjective evaluation method, and the analysis of the existing research results provides an area relative error criterion and is assisted by a pixel number error criterion to comprehensively evaluate the segmentation precision of the watershed algorithm image.
1) Area relative error criterion (precision factor: delta A )
In this embodiment, the relative error of the area is used as one of the criteria for evaluating the image segmentation accuracy, and the calculation method is as follows.
Let A 0 Representing the target volume area value in the reference data, A s Representing the target volume values in the segmented image results, their relative error delta A The method comprises the following steps:
wherein delta A And (5) evaluating the area precision factor of the image segmentation result. Obviously, delta A The smaller the size, the higher the segmentation accuracy, and conversely the worse.
2) Pixel number error criteria (precision factor: delta P )
The image segmentation accuracy is represented by dividing the number of erroneously segmented pixels by the total number of pixels, which is obtained by overlapping the reference image and the segmentation result, is consistent with the relative error criterion of the area, and is an accuracy evaluation performed from different angles.
Let P t Representing the number of correctly partitioned pixels, P w Representing the number of erroneously segmented pixels, the error rate delta P The method comprises the following steps:
wherein delta P Is the overall evaluation of the image segmentation accuracy, and obviously, delta P The smaller the size, the higher the segmentation accuracy, and conversely the worse. The evaluation criteria of the image segmentation accuracy of the watershed algorithm are shown in table 1.
Table 1 watershed algorithm image segmentation accuracy evaluation criterion
In order to evaluate the accuracy of the slope geological disaster boundary extraction method and the image segmentation extraction provided by the embodiment of the invention, an extraction test is carried out by using a data source of an experimental area.
The experimental area is located in Du Erping mining area peach blossom ditches in the mountain coal field of Taiyuan, shanxi province, china, and geological disasters such as unstable slopes, collapse and the like develop in the area, and various artificial buildings formed by artificial construction are formed. The geographical position of the experimental area is shown in fig. 1, and an unstable slope (marked with ∈ in fig. 1) in the selection area of the embodiment is used for carrying out a slope geological disaster boundary extraction experiment based on an area merging watershed algorithm for improving the Luv color space.
Data source and data preprocessing: and selecting the GF-2 remote sensing image as a data source, wherein the spatial resolution is 1m, and the imaging time is 2015. The geometric correction and orthographic correction of the image are selected from 1:10000 scale basic topographic map manufactured by aerophotogrammetry in 1999.
The image data preprocessing is completed by adopting geometric correction, image fusion, orthographic correction and cutting, and the preprocessed high-resolution remote sensing image: 1275×1503 pixels, as shown in fig. 2 a, are shown in fig. 3 for a field photograph of an unstable ramp 2017. And (3) performing visual interpretation on the test image by using the Arc GIS platform, and obtaining the accurate boundary of the target body as reference data after field verification and correction. As shown in fig. 4, a schematic diagram of experimental techniques of an embodiment is shown.
The slope geological disaster boundary extraction method for improving the watershed algorithm provided by the embodiment of the invention is realized based on an improved Luv color space region merging watershed algorithm (Luv-RMWS), belongs to a post-processing improvement algorithm, and is used for carrying out contrast enhancement pretreatment on a test image for comparison of segmentation efficiency and effect, wherein the processed image is shown as a B in a figure 2, and each segmentation method in the test adopts the same C value and D value, so that the comparison of test results is ensured. The original image shown in fig. 2 a and the contrast-enhanced image shown in fig. 2B are subjected to a region merging watershed algorithm segmentation test by adopting an RGB color space and a Luv color space respectively, the segmentation results are shown in fig. 5 and 6, and the statistical results of the segmentation process data are shown in table 2.
TABLE 2 statistics of image segmentation data
As can be seen from fig. 5, the segmentation result is improved compared with the original image after the contrast of the image is enhanced, but the vegetation region at the bottom of the slope still cannot be distinguished. As can be seen from fig. 6, the original image is severely under-segmented, the experimental extraction target object-unstable slope cannot be segmented, the segmentation result is good after the image contrast is enhanced, and the boundary consistency of the extracted unstable slope object is high. The statistical results in Table 2 show that the number of image spots segmented by the watershed algorithm is rapidly reduced after the contrast of the images by the RGB-RMWS method and the Luv-RMWS method is enhanced, the time consumption is slightly reduced, and the purposes of inhibiting over-segmentation and improving the efficiency are achieved. Meanwhile, the number of pattern spots and the time consumption after the RGB-RMWS method region combination are only half of that of the original image data, and the efficiency is greatly improved. The Luv-RMWS method increases the number of image spots after region merging compared with the original image data, but the time consumption is only half of that of the original image merging, and the time efficiency is greatly improved while a good segmentation result is obtained. Therefore, the improvement of the contrast enhancement preprocessing of the image has the obvious effect of improving the segmentation efficiency and the segmentation effect.
In order to select optimal segmentation and merging scale parameters, a multi-scale segmentation test is carried out on the test image, wherein C is respectively set as 100, 150, 200, … and 3000, and D is set as 100, 150, 200, … and 1000, so that a combination test is carried out. Four sets of test results, wherein C is 500, 1000, 1500, 2000, and D is 200, 300, 400, 500 are shown in FIG. 7.
Through a multi-scale Luv-RMWS method slope disaster boundary segmentation test, visual segmentation results and comparison can show that when the C value is gradually increased, the internal broken spots of the slope body are increased, and the slope body is over-segmented. And when D is gradually increased, the broken spots in the slope body are gradually reduced, and the extracted slope boundary is stabilized after D reaches 400. And the comparison of the experimental results is synthesized, when C is 500 and D is 400, the boundary segmentation effect of the target body is best, and the boundary is continuous, no broken spots exist in the spot body and the matching degree with the shape of the target body is high. Therefore, in the present embodiment, the optimal constant C in the minimum area determination threshold is set to 500, and the optimal value of the color difference value threshold D is set to 400. For different segmented images, trial and error can be performed to obtain the optimal values of constants C and D.
And (3) carrying out slope disaster boundary segmentation test on the RGB-RMWS method, and obtaining the following empirical values after visual segmentation results and comparison: the optimal constant C in the minimum area determination threshold is set to 100, and the optimal value of the color difference value threshold D is set to 11000.
The segmentation results of 100C and 11000D are taken as the segmentation results of the unstable slope body of the RGB-RMWS method, the segmentation results of 500C and 400D are taken as the segmentation results of the unstable slope body of the Luv-RMWS method, the two segmentation result data are converted into vector surfaces from grids, and the target body image spots are derived to be single surface files, reference data and test image superposition pairs, such as shown in figure 8.
The test results were analyzed as follows:
1) Time efficiency
The computer model used in the test is HP 2211f, and is specifically configured as follows: intel (R) Core (TM) i3 CPU, dominant frequency 3.20GHz;6.00GB of memory; a 64-bit operating system. The time used for the test is shown in Table 3 using the statistics of the programmed built-in timing variables.
Table 3 comparison table of the extraction time of unstable ramp boundaries for test images
As can be seen from Table 3, the RGB-RMWS method was used more often, reaching 182.209s, and the Luv-RMWS method was used only 39.702s. Thus, the Luv-RMWS method of improving the color space is significantly more time efficient than the unmodified RGB-RMWS method.
2) Extraction effect
(1) As can be seen from FIG. 8-A, the RGB-RMWS method has the best segmentation result, no broken spots exist in the image spots, but under-segmentation exists in the southwest part and the bottom of the slope of the image spots, and adhesion exists in the eastern boundary of the image spots.
(2) As can be seen from FIG. 8-B, the Luv-RMWS method has a better segmentation result, not only realizes better extraction of the target body, but also realizes combination of broken spots in the pattern spots, but excessive adhesion of the three large adhesion pattern spots into the pattern spots of the target body occurs in the north part and the southwest part of the pattern spots, and simultaneously under-segmentation phenomenon occurs in the south part of the pattern spots, namely the bottom of the target body, so that segmentation and combination of vegetation coverage areas at the bottom cannot be realized.
In summary, the RGB-RMWS method can obtain the optimal segmentation result of the target, although the visual effect is best, the time efficiency is low, the whole is under-segmented, while the Luv-RMWS method also has certain over-segmentation and under-segmentation phenomena, but the comprehensive segmentation efficiency and segmentation effect are good, especially the time efficiency for obtaining the optimal segmentation result of the target is far higher than that of the RGB-RMWS method, and the number of pattern spots in the result is relatively reasonable. Therefore, from the viewpoint of time efficiency and segmentation effect, the Luv-RMWS method comprehensively performs better than the RGB-RMWS method without color space conversion.
The area of the target object is extracted by reference data obtained by the extraction information of the test image and visual interpretation, and the RGB-RMWS method and the Luv-RMWS method are calculated by using the formula (13)Delta of (2) A . Meanwhile, dividing reference data obtained by extracting information of a test image and visual interpretation into two major types of target objects and non-target objects, taking the pixel number as a measuring unit, rasterizing the reference data of the target objects as a reference image, superposing the reference image with target object segmentation results of an RGB-RMWS method and a Luv-RMWS method to obtain correctly segmented pixels and incorrectly segmented pixel numbers, and calculating a segmentation result delta by using a formula (14) P . The calculation results are shown in Table 4.
Table 4 evaluation table for unstable ramp boundary extraction accuracy of test image
As can be seen from Table 4, delta of RGB-RMWS method A 6.21%, delta P 1.40% by weight, delta by the Luv-RMWS method A 4.92%, delta P 1.60%, it is apparent that the relative error of the area of the RGB-RMWS method is larger than that of the Luv-RMWS method, and the error of the number of pixels is smaller than that of the Luv-RMWS method, and the segmentation accuracy of the two methods is basically consistent.
1) Experiments show that the Luv-RMWS method has obvious advantages in information extraction time, avoids the complex process of post-treatment, and greatly improves the time efficiency. Meanwhile, the Luv-RMWS method automatically combines initial segmentation results, avoids subjectivity of manually combining and processing broken spots after classification, and has good comprehensive extraction efficiency and objectivity of extraction results.
2) The Luv-RMWS method does not need to establish a segmentation process rule, has simple merging process, is easy to understand the algorithm, and has strong objectivity and high reliability of segmentation results. The multi-scale segmentation test result of the unstable slope boundary of the test image shows that the Luv-RMWS method has good extraction reliability, object boundary fitness and extraction details. Meanwhile, the calculation result of the segmentation accuracy evaluation criterion shows that the extraction result of the Luv-RMWS method is higher in accuracy, and the evaluation result is consistent with the visual evaluation result, so that the reliability of the image segmentation accuracy evaluation criterion factor result of the watershed algorithm is shown.
The invention uses the Luv color space to locateThe Euclidean distance value chromatic aberration between the devices is used as a similarity measure, a homogeneity maximization criterion region merging algorithm is improved, a region merging watershed segmentation algorithm based on a Luv color space is provided, and through a trial-and-error multi-scale unstable slope boundary segmentation extraction experiment, the optimal segmentation and merging scale parameters of GF-2 images after the contrast enhancement of an experimental region are determined, wherein a minimum region A min The optimal threshold value C is determined to be 500, and the color difference value optimal threshold value D is determined to be 400.
2) By combining the existing research results and through comparative analysis, a watershed algorithm segmentation accuracy evaluation criterion system is established, and the method comprises the following steps: the improved area relative error criterion and pixel number error criterion are consistent with the comparison between the experimental result precision evaluation and the visual evaluation, and a new basis is provided for the image segmentation precision evaluation.
3) The post-improvement color space region merging watershed algorithm has good effect in slope disaster boundary extraction, obvious time efficiency advantage and reliable segmentation accuracy evaluation result, and has important application value in improving the slope boundary extraction efficiency.
Experimental results prove that the effectiveness of the improved algorithm for slope geological disaster boundary extraction provides new exploration for slope geological disaster information extraction based on remote sensing images, and objective and reliable data support is provided for post-disaster clear disaster-affected range and disaster relief emergency.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.
Claims (5)
1. The slope geological disaster boundary extraction method for improving the watershed algorithm is characterized by comprising the following steps of:
s1, dividing an image to be divided by a simulated immersion watershed algorithm; converting the RGB values of the pixels in each region obtained after the segmentation into Luv values, and calculating the Luv average value of all the pixels in each region to be used as the Luv value of the region;
s2, establishing four neighborhood arrays of each region with the Luv mean value as the region color, and simultaneously determining a minimum region judgment threshold value;
s3, scanning all the areas in sequence, judging whether the total number of pixels of the areas is smaller than a minimum area judging threshold value, and if so, classifying the pixels into minimum areas until all the minimum areas are found;
s4, traversing all adjacent areas of each determined minimum area, and calculating color difference values d of the minimum area and all the adjacent areas according to the Luv mean value of the adjacent areas i Satisfying the color difference value
Combining the region of (2) with the very small region; a new region is formed after combination, information of all adjacent regions of the new region after combination is refreshed, the mean value of the Luv values of the two regions before combination is used as the Luv value of the new region, and D is a color difference value threshold; color difference value d i The calculation formula of (2) is as follows:
wherein R is i |、|R j I respectively denote adjacent regions R i And a minimum region R j The number of pixels included in F c (R i )、F c (R j ) Respectively represent adjacent regions R i And a minimum region R j The average value of the colors in (1), n is the number of adjacent areas
S5, judging the pixel value of the new combined region, judging whether the new combined region is still a minimum region, and if so, returning to the step S4 to be combined again; if not, judging whether other minimum areas exist, if so, returning to the step S4 to continue merging until all the minimum areas are merged;
s6, converting the Luv values of all the combined areas into RGB values.
2. The method for extracting a slope geologic hazard boundary by improving a watershed algorithm according to claim 1, wherein in said step S2, a minimum area decision threshold a is determined min The values of (2) are:
A min ＝(M×N)/C；
wherein M is the row value of the image to be segmented, N is the column value of the image to be segmented, and C is a constant.
3. The method for extracting a slope geologic hazard boundary for improving a watershed algorithm according to claim 2, wherein the constant C has a value of 500 and the threshold value of the color difference has a value of D of 400.
4. The method for extracting the slope geologic hazard boundary for improving the watershed algorithm according to claim 1, wherein in the step S1, the image to be segmented is segmented by the watershed algorithm, specifically comprising the following steps:
(1) Converting pixel values of the image to be segmented to convert RGB values into gray values;
(2) Calculating to obtain gradients of each pixel point in the horizontal and vertical directions, and counting the frequency and the accumulation probability of each gradient;
(3) Sequencing according to the gradient values, determining the positions of the gradient values in a sequencing array, wherein the same gradient is in the same gradient level;
(4) Processing all pixel points of the first gradient level, checking whether the neighborhood of the point belongs to a certain region or watershed, if yes, adding the point into a first-in first-out queue;
(5) Starting to expand the existing basin according to the first-in first-out queue, scanning the pixel neighborhood in the first-in first-out queue, if the gradients are equal, namely the same gradient level is not a watershed, refreshing the identification of the pixel by using the identification of the neighborhood pixel, and circularly completing the expansion of all pixel points in the queue;
(6) Judging whether the pixel points are not identified, if so, continuing to execute the step (5) on the pixel points until all the pixel points in the queue are expanded;
(7) After the first gradient level is processed, returning to the step (4) to continuously process the next gradient level, and circulating until all gradient levels are processed, so as to obtain a water distribution line of the gradient image, namely a boundary line of image segmentation;
(8) The image is divided into a large number of areas by the dividing boundary line, and the areas are the image dividing results and represent different types of information.
5. The method for extracting a slope geologic hazard boundary for improving a watershed algorithm according to claim 1, wherein in the step S1, the process of converting RGB values of pixels in each region into Luv values is implemented based on color space conversion, and the conversion relationship is as follows:
u＝13L(u'-u n ')；
v＝13L(v'-v n ')；
wherein:
u'＝4X/(X+15Y+3Z)；
v'＝9Y/(X+15Y+3Z)；
u n '＝4X n /(X n +15Y n +3Z n )；
v n '＝9Y n /(X n +15Y n +3Z n )；
wherein, the L value represents the brightness of the pixel, the value range is 0-100, the u and v values represent the chromaticity coordinates, and the value range is-100 to the maximum100，X n ，Y n ，Z n Representing the coordinates of the CIE standard illuminant, X, Y, Z representing the values of the CIE XYZ color space.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911250891.5A CN111046783B (en) | 2019-12-09 | 2019-12-09 | Slope geological disaster boundary extraction method for improving watershed algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911250891.5A CN111046783B (en) | 2019-12-09 | 2019-12-09 | Slope geological disaster boundary extraction method for improving watershed algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111046783A CN111046783A (en) | 2020-04-21 |
| CN111046783B true CN111046783B (en) | 2023-06-30 |
Family
ID=70235096
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911250891.5A Active CN111046783B (en) | 2019-12-09 | 2019-12-09 | Slope geological disaster boundary extraction method for improving watershed algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111046783B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115689125B (en) * | 2022-12-27 | 2023-03-17 | 北京化工大学 | Multi-source data-based gas pollution source layout method |
| CN116597389B (en) * | 2023-07-18 | 2023-09-15 | 山东省地质测绘院 | Geological disaster monitoring and early warning method based on image processing |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103186904A (en) * | 2011-12-31 | 2013-07-03 | 北京新媒传信科技有限公司 | Method and device for extracting picture contours |
| CN107463991A (en) * | 2017-06-28 | 2017-12-12 | 西南石油大学 | A kind of Regional Landslide method for evaluating hazard based on slopes unit and machine learning |
| CN107945183A (en) * | 2017-06-28 | 2018-04-20 | 三亚中科遥感研究所 | A kind of combination improves the quick watershed segmentation methods for merging algorithm |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101923707B (en) * | 2009-07-23 | 2012-06-20 | 北京师范大学 | Watershed algorithm-based high spatial resolution multi-spectral remote sensing image segmentation method |
| US8478032B2 (en) * | 2011-05-24 | 2013-07-02 | Hewlett-Packard Development Company, L.P. | Segmenting an image |
| CN102509097B (en) * | 2011-09-29 | 2013-10-23 | 北京新媒传信科技有限公司 | Method and device for image segmentation |
| CN102999888B (en) * | 2012-11-27 | 2015-02-25 | 西安交通大学 | Depth map denoising method based on color image segmentation |
| CN104881865B (en) * | 2015-04-29 | 2017-11-24 | 北京林业大学 | Forest pest and disease monitoring method for early warning and its system based on unmanned plane graphical analysis |
| CN105844292B (en) * | 2016-03-18 | 2018-11-30 | 南京邮电大学 | A kind of image scene mask method based on condition random field and secondary dictionary learning |
| CN107103317A (en) * | 2017-04-12 | 2017-08-29 | 湖南源信光电科技股份有限公司 | Fuzzy license plate image recognition algorithm based on image co-registration and blind deconvolution |
| CN109741337B (en) * | 2018-12-11 | 2022-11-29 | 太原理工大学 | Region merging watershed color remote sensing image segmentation method based on Lab color space |
-
2019
- 2019-12-09 CN CN201911250891.5A patent/CN111046783B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103186904A (en) * | 2011-12-31 | 2013-07-03 | 北京新媒传信科技有限公司 | Method and device for extracting picture contours |
| CN107463991A (en) * | 2017-06-28 | 2017-12-12 | 西南石油大学 | A kind of Regional Landslide method for evaluating hazard based on slopes unit and machine learning |
| CN107945183A (en) * | 2017-06-28 | 2018-04-20 | 三亚中科遥感研究所 | A kind of combination improves the quick watershed segmentation methods for merging algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111046783A (en) | 2020-04-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112419250B (en) | Pavement crack digital image extraction, crack repair and crack parameter calculation method | |
| CN107610114B (en) | optical satellite remote sensing image cloud and snow fog detection method based on support vector machine | |
| CN110309781B (en) | House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion | |
| CN109242870A (en) | A kind of sea horizon detection method divided based on image with textural characteristics | |
| CN106530313B (en) | A kind of real-time detection method for sea sky based on region segmentation | |
| CN113554629A (en) | Strip steel red rust defect detection method based on artificial intelligence | |
| CN111046783B (en) | Slope geological disaster boundary extraction method for improving watershed algorithm | |
| CN103957771A (en) | Image processing device, image processing method, and image processing program | |
| US8559714B2 (en) | Post processing for improved generation of intrinsic images | |
| CN111046884B (en) | Slope geological disaster extraction method of multi-feature auxiliary watershed algorithm | |
| US20140294296A1 (en) | Spatially varying log-chromaticity normals for use in an image process | |
| CN113033315A (en) | Rare earth mining high-resolution image identification and positioning method | |
| CN115641327A (en) | Building engineering quality supervision and early warning system based on big data | |
| CN111008664A (en) | Hyperspectral sea ice detection method based on space-spectrum combined characteristics | |
| CN115272353A (en) | Image processing method suitable for crack detection | |
| CN109741337B (en) | Region merging watershed color remote sensing image segmentation method based on Lab color space | |
| CN111091071A (en) | Underground target detection method and system based on ground penetrating radar hyperbolic wave fitting | |
| CN106204596B (en) | Panchromatic waveband remote sensing image cloud detection method based on Gaussian fitting function and fuzzy mixed estimation | |
| US8428352B1 (en) | Post processing for improved generation of intrinsic images | |
| CN113643312B (en) | Cloud layer segmentation method based on true color satellite cloud picture and image processing | |
| EP2776979B1 (en) | Post processing for improved generation of intrinsic images | |
| CN115984360A (en) | Method and system for calculating length of dry beach based on image processing | |
| CN114862883A (en) | Target edge extraction method, image segmentation method and system | |
| CN104504690B (en) | A kind of object support relationship estimating method based on Kinect | |
| CN116563659A (en) | Optical smoke detection method combining priori knowledge and feature classification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |